Skip to content
All insights
Guide · Strategy

How to Staff AI Without Hiring a Data Scientist

The hire that decides whether an AI system is alive in a year is not a machine learning engineer. It is the person inside your company whose name is on the runbook.

· 4 min read

Patrick Laughlin, Founder, Chief AI Officer

Naval Academy, class of 2011. Decorated naval aviator, 10+ years. President, 2023 Inc. 5000 company.

An early question on scoping calls is whether the company needs to hire a data scientist. For most companies between two million and five hundred million dollars in revenue, the answer is no.

You are not training models. You are wiring models that already exist into work that already happens. That is an integration and process problem with a little prompt engineering on top, and the people who are good at it are engineers and operators, not researchers.

What you do need is a person who can move data, a person who can build, and one person inside your company whose name is on the system. The third one gets skipped, and the third one decides whether any of it is alive in a year.

What a named owner looks like

A retirement planning firm of a few dozen people has ten agents running across eight systems with us, starting with an intake agent built in the first week.

Ten agents in a firm that size is a lot of surface area. What holds it together is their chief technology officer, who asked for most of it. He sits in the demos. He tells us when something is wrong, in specifics, the same week. He has the marketing portal open on his home screen.

He does not write code and has no plans to. He owns the systems, which is a different thing and a rarer one.

What you have to give that person

Naming an owner and giving them nothing is worse than not naming one, because now there is someone to blame.

A build they can read. Where the work allows it, we build the automation layer in a visual workflow tool running on the client's own server. Someone who is not a developer can open a step, read the prompt, change the wording, and save. That sounds minor. It is the difference between a change and a ticket. A firm that can adjust its own prompts stops needing us for the small things within a couple of months, which is the point.

The credentials. Their accounts. Their server. Their keys. At any point they could walk away and everything I worked on is still running, on infrastructure they control. If that is not true, you do not have an owner, you have a contact at a vendor.

A runbook. What it does, when it runs, what it costs to run, what it looks like when it breaks, and who to call. Four pages, not forty.

Hours. An owner with no time is a title. Put it in their week.

What training does and does not do

I ran a company-wide session for a natural skincare company, a couple of dozen people in the room, after we had done a full discovery and our first builds there. It was never going to turn the room into expert prompt writers, and it did not try.

What came out was that, in the weeks after, most of the team settled on one assistant on their own instead of arguing about five, and a handful of people left with a specific use for it inside their own week. The training post tells that story in full. For a single session, that is a good result and it is the result I would promise. Anyone selling you a workshop that turns your staff into AI practitioners has not run one.

Training moves the ceiling on what people will accept. Systems move the floor on what they have to do. You want both, and you should not confuse them.

The hires that are sometimes real

A data engineer, eventually, if your operating data lives in six systems that disagree and you intend to run reporting off it. That is a real role, and it is usually the first technical hire that pays for itself.

An analyst who owns definitions, if nobody currently owns what a number means.

Almost nobody I work with needs a machine learning engineer. If you do, you will know, because you will have a proprietary dataset and a modeling problem rather than a workflow problem.

What I will not do

I will not leave a build with no name attached to it. If a company cannot tell me who owns a system once it is live, that conversation is worth more than the build, and I would rather have it before the invoice than a year later when nobody can log in.

Tell us what you want to get done.

A first conversation with a Chief AI Officer about your goals and what you need, whether that is a build, training, change management or expert advice. We shape the engagement around your answer.

A developer at work on a build